The pretesting effect: wrong guesses that help

Memory researchers call it the pretesting effect, or sometimes "errorful generation": being tested on material before you've learned it improves how well you retain it afterward, compared with simply studying the correct answer from the start. A major 2023 review by Steven Pan and Shana Carpenter pulled this literature together, and the pattern is consistent — the act of reaching for an answer, even an answer you don't have yet, prepares your memory to grab the real one when it arrives.

Two details matter for language learners. First, the effect doesn't depend on guessing right — a wrong guess followed by immediate feedback still beats passive study. Second, this isn't just a lab curiosity with English trivia: a 2026 study out of the National University of Singapore found the effect holds for second-language vocabulary learned in app-style tasks — guess first, then see the meaning — which is exactly the situation a Mandarin learner is in.

The intuition, roughly: when you guess what 你好 (nǐ hǎo) means before being told, you activate everything you might know that's relevant, notice the gap where the answer should be, and become briefly, genuinely curious. The correct answer then lands in prepared ground instead of sliding past.

Confident mistakes are gold: the hypercorrection effect

The second finding is stranger. You'd expect your most confident errors — the ones where you were sure and still wrong — to be the hardest to fix. The opposite is true. Corrections to high-confidence errors are remembered better than corrections to timid ones. Researchers call this the hypercorrection effect, and the mechanism seems to be surprise: being confidently wrong grabs your attention in a way a hesitant miss never does, and attention is what memory runs on.

There's an important footnote, though. Follow-up studies found that hypercorrection is strongest in the first days after the correction — and that around the one-week mark, the original confident error has a tendency to creep back. The practical implication is very specific: a confidently-missed word doesn't just need correcting once, it needs a deliberate second look about a week later, right when the old wrong answer would otherwise resurface.

Productive failure: attempting beats being told

The third line of research zooms out from single words to whole learning designs. A 2021 meta-analysis by Tanmay Sinha and Manu Kapur — synthesizing 166 comparisons — found that learners who attempt problems before receiving instruction reliably outperform learners who receive instruction first. The approach is called productive failure, and the effect held moderately-to-strongly when the method was implemented faithfully.

So why doesn't every learning app work this way? Because failing feels bad. A quiz you haven't been prepared for reads as unfair, and consumer apps optimize hard for users never feeling stupid. The research is clear that the cognition of guess-first is a win; the reason it stays out of apps is the emotion. Which means the design problem isn't "should learners guess first" — it's "how do you make guessing first feel safe."

Why most apps don't use any of this

Look at how mainstream tools actually handle your answers. Flashcard systems like Anki ask you to grade yourself ("again / hard / good / easy") — powerful, but entirely manual, and with no pedagogy wrapped around your errors. Most popular learning apps track exactly one bit of information per answer: right or wrong. A confident wrong answer and a wild guess that happened to miss are treated identically; a hard-won correct answer and an instant, fluent one are treated identically too. All of the signal the research above runs on — how sure you were — is thrown away.

How CiaoSpeak turns the research into a system

We built Hunches as a direct implementation of these three findings, with the emotional problem solved by framing:

  • Guess first, zero stakes. A new word opens with "What's your hunch?" — the character, its audio, and four possible meanings — before the teaching happens. Guessing is free by design: no XP is lost, nothing goes on your mistake record, and the feedback is warm whichever way it goes. Wrong hunches aren't failures; per the pretesting research, they're the mechanism.
  • Confidence is read, never asked. Instead of interrupting you with "how sure were you?" taps, CiaoSpeak quietly compares your response time on graded exercises against your own typical pace. There's no visible timer and no pressure — just the signal, recovered.
  • Surprises get celebrated — then ambushed. A fast, confident, wrong answer is the hypercorrection moment, so the app treats it as a find, not a failure — and then schedules that word's review to land about a week later, exactly where the research says corrected confident errors start slipping back. It may even resurface casually in your AI conversation practice, as an ordinary question you don't know is a check.
  • Calibration you can watch. Over time, the app shows you how often you're right when you feel sure — a skill called calibration that most learners never get to see about themselves.

The honest fine print

Each mechanism above is well-evidenced on its own — pretesting, hypercorrection, and productive failure all have real literatures behind them. Combining them into one system for Mandarin vocabulary is our synthesis, and we're validating that combination with real learners rather than claiming numbers we haven't measured. You'll notice this page says "measurably helps it stick," not "triples your learning speed" — that's deliberate.

And guessing-first is one piece, not the whole method: sounds still come before everything, which is why CiaoSpeak's very first lesson trains your ear on the four tones before any vocabulary at all.